Overview

For creating a pretty map layout, R may be somewhat limited compared to ArcMap, especially since ArcMap has a graphical interface designed specifically for making nice layouts. When we get to spatial analysis, we can do mamy things in R that we will learn how to do in ArcMap.

Some notes: The sf package is a powerful and relatively new R package that works with vector spatial data (polygons, lines, points - as opposed to rasters). It is designed to work well within the tidyverse functions and workflow. An older system of spatial analysis was built around a package called sp, and its cousins rgdal and rgeos, which implement a couple of general spatial analysis libraries: GDAL, or Geospatial Data Abstraction Library and GEOS, or Geometry Engine - Open Source. The sf package relies on the same libraries but is much less annoying than the older sp package.

Load the spatial data

The .gdb file extension indicates a geodatabase, usually with multiple layers and attributes. If these come from ArcMap, they may have some proprietary features that makes them more difficult to work with. Another commonly used format is the shapefile (.shp, which is always accompanied by a few related files, including a projection, .prj, and an attribute table, .dbf).

For this assignment, the HW1.gdb file is a geodatabase containing just a grid of the world cut into 30° squares. The CountryWatch.lyr combines a world map with country attributes and even some formatting. But it’s a pain to work with outside of ArcMap. Instead, I used ArcMap to load the CountryWatch.lyr and saved it back out as a shapefile (.shp with the spatial information as a polygon feature for each country) with its attribute table (.dbf with all CountryWatch data tied to each feature in the shapefiles).

In the sf package, most functions start with st_ (for “spatiotemporal”), including st_read and st_write, but there’s also read_sf and write_sf which are basically the same with a few simplifying assumptions. Point the read functions toward the shapefile .shp, and it will automatically pull in the .dbf and .prj and whatnot.

country_data_sf <- st_read('shapefiles/Countries.shp') %>%
  janitor::clean_names()
## Reading layer `Countries' from data source `/home/ohara/github/esm263/HW1/shapefiles/Countries.shp' using driver `ESRI Shapefile'
## Simple feature collection with 250 features and 61 fields
## geometry type:  MULTIPOLYGON
## dimension:      XY
## bbox:           xmin: -180 ymin: -90 xmax: 180 ymax: 83.6236
## epsg (SRID):    4326
## proj4string:    +proj=longlat +datum=WGS84 +no_defs
grid_sf <- read_sf('shapefiles/World30.shp') %>%
  janitor::clean_names()

names(country_data_sf)
##  [1] "objectid"   "fips_cntry" "gmi_cntry"  "iso_2digit" "iso_3digit"
##  [6] "cntry_name" "long_name"  "sovereign"  "pop_cntry"  "curr_type" 
## [11] "curr_code"  "landlocked" "sqkm"       "sqmi"       "colormap"  
## [16] "shape_leng" "shape_area" "objectid_1" "fips_cnt_1" "cntry_na_1"
## [21] "politi_map" "cw_data"    "sqkm_1"     "coast_km"   "totpop00"  
## [26] "grwthrat00" "perurban00" "popdens00"  "per0_14_00" "per15_5900"
## [31] "per60_up00" "depratio00" "fertrate99" "brthrate00" "net_migr00"
## [36] "deathrat00" "life_exp00" "femlfexp00" "mallfexp00" "adj_exp99" 
## [41] "inf_mort00" "sexratio00" "lit_rate00" "edu_levl95" "gdp_99"    
## [46] "gdpgrwrt99" "gdp_pcap99" "gdp_pcgr99" "imports_99" "exports_99"
## [51] "engconpc97" "coal_res98" "coalprod98" "coal_con98" "pet_prod98"
## [56] "pet_con98"  "ngasprod98" "ngas_con98" "elec_con98" "irnoprod98"
## [61] "ggas_ems98" "geometry"

viewing data from an sf object

A simple features object is essentially a dataframe with a row for each feature (e.g. country) and columns for all the variables, and one extra column, a geometry column, that holds all the spatial info for that feature. Using View() on an sf object bogs things down since it tries to show that geometry column.

You can separate the data from the geometry to make the data part a little easier to work with, though. Note I sometimes use a tag at the end of the object name so I can easily see what type: _sf for simple features, _df for dataframes, _rast for rasters, etc. Not necessary but I find it helpful. The geometry column is “sticky” in an sf object, so select can’t drop it - so instead we coerce the object to a data frame, then select out the geometry…

country_data_df <- country_data_sf %>%
  as.data.frame() %>%
  select(-geometry)

making a very simple map

The ggplot2 package works well with simple features, using the geom_sf() layer. You can put the simple features data into the initial ggplot call, but because it’s an odd data type, and we’re often using multiple sf layers, it can simplify things to just call it and set the aesthetics in the geom_sf(data = ...) call.

We need to fix the NA values first though. I’m mapping the Dependency Ratio so I’ll just select that column (since geometry is “sticky”, so I don’t need to worry about it in my select call here - it will stay with the object).

depratio_sf <- country_data_sf %>%
  select(depratio = depratio00) %>%
  mutate(depratio = ifelse(depratio < 0, NA, depratio))

ggplot() +
  geom_sf(data = depratio_sf, aes(fill = depratio)) +
  scale_fill_distiller(palette = 'YlOrBr', 
    ### see RColorBrewer::display.brewer.all() for different palettes
                       direction = 1,
                       limits = c(0, NA)) 

    ### sets bottom limit to 0, top limit to infinity, forcing the 
    ### palette to go all the way to zero

Note that we don’t even need the grid from the World30 shapefile, ggplot adds graticules for us. Note also that the color scale is continuous - not broken into quintiles or anything.

Making it fancy

With a basic map, we can start to use theme and other plot options to make it look a little nicer. In the assignment, I used a discrete scale for colors, based on certain cut points (60%, 75%, 90%, 100%, more than 100%). Here let’s try dividing into quantiles - I like quintiles, so the bottom 20%, second 20%, etc. Now we’re mapping a discrete variable (quantiles) instead of a continuous (dependency ratio), so we’ll use scale_fill_brewer() which works for discrete.

Note on the mutate stuff: first calc which quintile each country falls in, then give it a text description. We can group by quintile and pull the min/max for the text description. Since we arranged by dependency ratio, we know the quintiles are in order, so fct_inorder forces the text descriptions to be factors in the correct order.

dep_quintiles_sf <- depratio_sf %>%
  arrange(depratio) %>% 
  mutate(dep_qtile = ntile(depratio, 5)) %>%
  group_by(dep_qtile) %>%
  ### note: to escape a percent sign in sprintf, use %% - totally not confusing
  mutate(qtile_desc = sprintf('%s%% - %s%%', round(min(depratio)), round(max(depratio))),
         qtile_desc = ifelse(is.na(depratio), NA, qtile_desc), ### for NAs
         qtile_desc = fct_inorder(qtile_desc))

### to check what the data looks like now
# dep_qtiles_df <- dep_quintiles_sf %>%
#   as.data.frame() %>%
#   select(-geometry)

ggplot() +
  geom_sf(data = dep_quintiles_sf, aes(fill = qtile_desc)) +
  scale_fill_brewer(palette = 'YlOrBr', 
                    direction = 1,
                    na.value = 'grey50')

Now let’s change the background to blue, trim the map edges in, add a title, subtitle, and caption. We’ll also drop the degree markers around the edges. The theme ggplot layer can change the format of pretty much any element you want to change.

We’ll store this plot in an object, so it doesn’t immediately print to the screen; instead we’ll ggsave it and then include the saved plot instead.

nicer_plot <- ggplot() +
  geom_sf(data = dep_quintiles_sf, aes(fill = qtile_desc),
          color = 'grey20', size = .1) +
  scale_fill_brewer(palette = 'YlOrBr', 
                    direction = 1,
                    na.value = 'grey85') +
  theme(text = element_text(size = 6), ### default text
        plot.title = element_text(face = 'bold.italic', size = 12),
        axis.text = element_blank(),
        axis.ticks = element_blank(),
        legend.key.size = unit(3, 'mm'),
        legend.position = c(.08, .22),
        legend.box.background = element_rect(color = 'grey20', size = .25),
        panel.background = element_rect(fill = 'azure1', color = 'grey50')) +
  coord_sf(expand = FALSE, datum = NA) + ### datum = NA suppresses the graticules
  labs(fill =  'Dependency ratio',
       title = 'Dependency Ratio (2000 estimates)',
       caption = paste("Map drawn by Casey O'Hara, 1/20/20",
                       "Source: Bureau of Census etc.",
                       "CountryWatch and ESRI, etc.",
                       sep = '\n')) ### use \n to get multiple lines

ggsave('basic_plot.png', dpi = 300, width = 6.5, height = 3.5)

Replicating the white-on-black map

coal_sf <- country_data_sf %>%
  select(coal_res = coal_res98) %>%
  mutate(coal_res = ifelse(coal_res < 0, NA, coal_res)) %>%
  arrange(coal_res) %>% 
  mutate(coal_cat = case_when(coal_res <= 5000   ~ '0-5B tons',
                              coal_res <= 25000  ~ '5B-25B tons',
                              coal_res <= 100000 ~ '25B-100B tons',
                              coal_res <= 200000 ~ '100B-200B tons',
                              coal_res > 200000  ~ '200B+ tons',
                              TRUE ~ 'No Data'),
         coal_cat = fct_inorder(coal_cat))

We’ll use the coal_cat variable to drive fill (shades of grey with one more saturated for emphasis), color (all grey except emphasis), and linetype (solid for all but NA). Then we can use scale_X_manual() to set scales for fill, color, and linetype separately.

coal_lvls <- levels(coal_sf$coal_cat)
# "0-5B tons"      "5B-25B tons"  "25B-100B tons"  
# "100B-200B tons" "200B+ tons"   "No Data"
coal_fills <- c('grey20', 'grey40', 'grey60', 'grey80', 'grey95', 'black')
coal_cols  <- c('grey50', 'grey50', 'grey50', 'grey50', '#ffff44', '#882222')

inverse_plot <- ggplot() +
  geom_sf(data = coal_sf, 
          aes(fill = coal_cat, color = coal_cat),
          size = .15) +
  scale_fill_manual(breaks = coal_lvls, values = coal_fills) +
  scale_color_manual(breaks = coal_lvls, values = coal_cols) +
  theme(text = element_text(size = 6, color = 'white', family = 'AvantGarde'), 
          ### default text; note AvantGarde might not work on some systems
        plot.title = element_text(face = 'bold', size = 12),
        plot.subtitle = element_text(face = 'bold', size = 9),
        panel.background = element_rect(fill = 'black'),
        panel.border = element_rect(color = 'grey30', fill = NA, size = 1),
        panel.grid   = element_line(color = 'grey30', size = .25),
        axis.text = element_blank(),
        axis.ticks = element_blank(),
        legend.key.size = unit(3, 'mm'),
        legend.key = element_rect(color = NA, fill = NA),
        legend.position = c(.08, .25),
        legend.background = element_rect(fill = NA),
        plot.background = element_rect(color = 'grey50', fill = 'black', size = 1)) +
  coord_sf(expand = FALSE) +
  labs(fill =  'Coal reserves', color = 'Coal reserves',
       title = 'NATIONAL COAL RESERVES 1998',
       caption = paste("Map drawn by Casey O'Hara, 1/20/20",
                       "Source: Bureau of Census etc.",
                       "CountryWatch and ESRI, etc.",
                       sep = '\n')) ### use \n to get multiple lines

ggsave('inverse_plot.png', dpi = 300, width = 6.5, height = 3.5)

Multiple maps and insets with cowplot

The cowplot package allows you to combine multiple plots/maps/graphics objects (grobs) into one. For example, a figure in a journal article might have parts A, B, and C; cowplot can do that. Here we’ll do something simple with it - first we’ll just put both into a single figure, using the cowplot::plot_grid() function.

Note that you might have to play with the saved image size to make it look good. Because geom_sf locks the aspect ratio (height vs width) it might add a lot of white space to make it fit your saved plot size…

multi_plot <- cowplot::plot_grid(nicer_plot, inverse_plot, 
                                 align = 'h', ncol = 1, labels = c('A', 'B'))

ggsave('multi_plot.png', width = 6.5, height = 7)

knitr::include_graphics('multi_plot.png') ### a different way of including images in R Markdown

Adding an inset

Now let’s add one of our previous maps (nicer_plot) to the other one (inverse_plot) as an inset. Here we’ll use cowplot::ggdraw() which sort of sets up a combo plot canvas to which we can add plots, and cowplot::draw_plot() to add plots to it. By default, cowplot thinks of the canvas as an x and y axis going from 0 (bottom left corner) to 1 (moving up or to the right), so we can position and size our plots using this.

We’ll deconstruct our maps to separate the map itself from the legend using cowplot::get_panel and cowplot::get_legend (note, this also drops the title etc). The legend objects don’t play all that well here, so there are some alignment issues, but you can get the idea.

### Modifying the legend so it's black text on white background, and matches
### the legend for the dependency ratio plot...
inv_plot_new_legend <- inverse_plot + 
  theme(legend.background = element_rect(fill = 'white'),
        legend.box.background = element_rect(color = 'grey20', size = .25),
        text = element_text(color = 'black', family = 'sans'))

coal_map <- get_panel(inverse_plot)
coal_legend <- get_legend(inv_plot_new_legend)
depratio_map <- get_panel(nicer_plot)
depratio_legend <- get_legend(nicer_plot)

inset_plot <- ggdraw() +
  draw_plot(coal_map, 
            x = 0, y = .1,              ### locate the bottom left corner at (0, .1)
            width = 1, height = .9) +   ### make the plot 1 wide (full panel) and .9 tall (from 0.1 to 1)
  draw_plot(depratio_map,
            x = .04, y = .35,            ### locate the bottom left at (0.04, 0.30)
            width = .20, height = .20) + ### shrink this map to be smaller.
  draw_plot(depratio_legend,
            x = .06, y = .10,            ### locate the bottom left at (.06, .05)
            width = .2, height = .3) +   ### legend interacts weirdly with size and location so play around
  draw_plot(coal_legend,
            x = .76, y = .10,            ### locate the bottom left at (.06, .05)
            width = .2, height = .3) 

ggsave('inset_plot.png', width = 6.5, height = 4)

knitr::include_graphics('inset_plot.png')

more complicated example

Not gonna put the code here for this one, but to give a sense of what you can do with cowplot and ggplot, here’s a complex map graphic I assembled for a paper, Changes in ocean health in British Columbia from 2001 to 2016